new england

The
journal of medicine
established in 1812 August 22, 2019 vol. 381  no. 8

Ambient Particulate Air Pollution and Daily Mortality in 652 Cities

C. Liu, R. Chen, F. Sera, A.M. Vicedo‑Cabrera, Y. Guo, S. Tong, M.S.Z.S. Coelho, P.H.N. Saldiva, E. Lavigne,
P. Matus, N. Valdes Ortega, S. Osorio Garcia, M. Pascal, M. Stafoggia, M. Scortichini, M. Hashizume, Y. Honda,
M. Hurtado‑Díaz, J. Cruz, B. Nunes, J.P. Teixeira, H. Kim, A. Tobias, C. Íñiguez, B. Forsberg, C. Åström,
M.S. Ragettli, Y.-L. Guo, B.-Y. Chen, M.L. Bell, C.Y. Wright, N. Scovronick, R.M. Garland, A. Milojevic, J. Kyselý,
A. Urban, H. Orru, E. Indermitte, J.J.K. Jaakkola, N.R.I. Ryti, K. Katsouyanni, A. Analitis, A. Zanobetti, J. Schwartz,
J. Chen, T. Wu, A. Cohen, A. Gasparrini, and H. Kan​​

a bs t r ac t

BACKGROUND
The systematic evaluation of the results of time-series studies of air pollution is challenged The authors’ full names, academic de-
by differences in model specification and publication bias. grees, and affiliations are listed in the
Appendix. Address reprint requests to Dr.
Kan at P.O. Box 249, 130 Dong-An Road,
METHODS Shanghai 200032, China, or at ­ kanh@​
We evaluated the associations of inhalable particulate matter (PM) with an aerodynamic ­fudan​.­edu​.­cn.
diameter of 10 μm or less (PM10) and fine PM with an aerodynamic diameter of 2.5 μm Drs. Liu and R. Chen and Drs. Gasparrini
or less (PM2.5) with daily all-cause, cardiovascular, and respiratory mortality across multi- and Kan contributed equally to this article.
ple countries or regions. Daily data on mortality and air pollution were collected from 652 N Engl J Med 2019;381:705-15.
cities in 24 countries or regions. We used overdispersed generalized additive models with DOI: 10.1056/NEJMoa1817364
random-effects meta-analysis to investigate the associations. Two-pollutant models were Copyright © 2019 Massachusetts Medical Society.

fitted to test the robustness of the associations. Concentration–response curves from each
city were pooled to allow global estimates to be derived.
RESULTS
On average, an increase of 10 μg per cubic meter in the 2-day moving average of PM10
concentration, which represents the average over the current and previous day, was associ-
ated with increases of 0.44% (95% confidence interval [CI], 0.39 to 0.50) in daily all-cause
mortality, 0.36% (95% CI, 0.30 to 0.43) in daily cardiovascular mortality, and 0.47% (95%
CI, 0.35 to 0.58) in daily respiratory mortality. The corresponding increases in daily mortal-
ity for the same change in PM2.5 concentration were 0.68% (95% CI, 0.59 to 0.77), 0.55%
(95% CI, 0.45 to 0.66), and 0.74% (95% CI, 0.53 to 0.95). These associations remained
significant after adjustment for gaseous pollutants. Associations were stronger in locations
with lower annual mean PM concentrations and higher annual mean temperatures. The
pooled concentration–response curves showed a consistent increase in daily mortality with
increasing PM concentration, with steeper slopes at lower PM concentrations.
CONCLUSIONS
Our data show independent associations between short-term exposure to PM10 and PM2.5
and daily all-cause, cardiovascular, and respiratory mortality in more than 600 cities across
the globe. These data reinforce the evidence of a link between mortality and PM concentra-
tion established in regional and local studies. (Funded by the National Natural Science
Foundation of China and others.)

n engl j med 381;8 nejm.org  August 22, 2019 705

The New England Journal of Medicine
Downloaded from nejm.org by Javier de Jesus Oquendo Contreras on September 10, 2019. For personal use only. No other uses without permission.
Copyright © 2019 Massachusetts Medical Society. All rights reserved.
 The n e w e ng l a n d j o u r na l
T
he adverse health effects of short-
term exposure to ambient air pollution are
well documented.1-3 Particulate matter (PM),
especially, arouses public health concerns be-
A Quick Take is
cause of its toxicity and the widespread human
available at exposure to this pollutant. PM, which includes
NEJM.org inhalable particles with an aerodynamic diameter
of 10 μm or less (PM10) and fine particles with
an aerodynamic diameter of 2.5 μm or less
(PM2.5), is emitted from combustion sources or
formed through atmospheric chemical transfor-
mation. Given the extensive evidence regarding
their effects of health, the daily and annual
mean concentrations of PM10 and PM2.5 are regu-
lated according to the World Health Organiza-
tion (WHO) Air Quality Guidelines4 and standards
in major countries.
Numerous time-series studies have examined
the associations between short-term PM expo-
sures and daily mortality.5-9 However, most evi-
dence has been obtained from studies in single
cities, regions, or countries, and there are chal-
lenges in comparing these results and in synthe-
sizing effect estimates because of different mod-
eling approaches and potential publication bias.
These limitations can be addressed by perform-
ing international, multicenter studies that adopt
the same analytic protocol and model specifica-
tions to estimate globally representative asso-
ciations of PM10 and PM2.5 exposures with daily
mortality. We established the Multi-City Multi-
Country (MCC) Collaborative Research Network
to perform a global assessment of the effects of
weather or climate on mortality.10,11 This network
allowed us to examine and compare the associa-
tions of PM concentrations with daily all-cause,
cardiovascular, and respiratory mortality at the
global, regional, and country level with the use of
a standardized analytic framework.
Me thods
Data Collection
We obtained health and environmental data from
the MCC database, which has been described
previously.10,12 The current analysis was limited
to locations that had available data on air pollu-
tion (652 urban areas in 24 countries or regions,
with the data covering the period from 1986
through 2015) (Table S1 in the Supplementary
Appendix, available with the full text of this arti-
706 n engl j med 381;8 nejm.org  August 22, 2019
The New England Journal of Medicine
Downloaded from nejm.org by Javier de Jesus Oquendo Contreras on September 10, 2019. For personal use only. No other uses without permission.
Copyright © 2019 Massachusetts Medical Society. All rights reserved.
of m e dic i n e
cle at NEJM.org). Data on mortality were obtained
from local authorities within each country. Causes
of death were classified according to codes in the
International Classification of Diseases, 9th Revision
(ICD-9) or 10th Revision (ICD-10), whichever was
available. In each location, mortality was repre-
sented by daily counts of either death from non-
external causes (ICD-9 codes 0 to 799 and ICD-10
codes A0 to R99) or, when such data were un-
available, daily counts of death from any cause.
We also collected mortality data for two main
causes of death: cardiovascular disease (ICD-10
codes I00 to I99) and respiratory disease (ICD-10
codes J00 to J99).13
We obtained daily data on PM10 in 598 cities
and on PM2.5 in 499 cities. Data on both pollutants
were available in 445 cities in 16 countries or re-
gions. The geographic distributions of the cities
that had data on PM10 and PM2.5, as well as the
annual mean PM concentrations over the period
studied for each city, are provided in Figure 1 and
Figure 2, respectively (also see the interactive map,
available at NEJM.org). Daily data on gaseous pol-
lutants (ozone, nitrogen dioxide, sulfur dioxide,
and carbon monoxide) were obtained where avail-
able. We also collected data on the daily mean
temperature and daily mean relative humidity. To
avoid potential consequences of including outly-
ing values of exposure data, we used trimmed
data, in which the highest 5% and lowest 5% of
PM10 and PM2.5 measurements were excluded.14
Statistical Analysis
The associations of PM10 and PM2.5 concentrations
with daily all-cause, cardiovascular, and respira-
tory mortality were assessed in separate analyses
with the use of a standard time-series approach.
We followed a two-stage analytic protocol, which
had been developed and widely applied in previ-
ous multicity time-series studies.15,16
In the first stage, we estimated city-specific
associations of PM concentration with mortality
using quasi-Poisson generalized additive models.
In accordance with the approaches used in previ-
ous studies,16,17 the following covariates were in-
cluded in the main model: a natural cubic smooth
function with 7 degrees of freedom (df) per year
to control for underlying time trends in mortal-
ity, an indicator day-of-week variable to account
for short-term weekly variations, and natural
spline functions with 6 df for temperature and 3 df
 Air Pollution and Mortality in 652 Cities

Mean Annual PM10

Concentration (µg/m3)
0–10
11–20
21–30
31–40
41–60
61–80
81–100
101–150
151–200
>200

Kilometers
0 2500 5000 10,000

Figure 1. Distribution of the Cities with Data on PM10.

Shown is the geographic distribution of the 598 cities in the 24 countries and regions that had available data on particulate matter with
an aerodynamic diameter of 10 μm or less (PM10) and were included in the analysis. Also shown are the annual mean PM10 concentrations.
See the interactive map, available at NEJM.org.

for relative humidity to control for potentially
nonlinear confounding effects of weather condi-
tions in areas where such data were available. To
determine an appropriate lag time (i.e., the num-
ber of days between exposure and the estimated
effect) for PM and temperature to be used in the
main analyses, we compared a variety of lag days
using generalized cross-validation scores.
In the second stage, we used random-effects
models to pool the estimates of the city-specific
associations of PM concentrations with mortality.18
We then reported the pooled estimate and re-
lated 95% confidence intervals as the percentage
change in daily mortality per 10-μg-per-cubic-
meter increase in PM concentrations. Between-
city heterogeneity was quantified with the use of
the I2 statistic.
In addition to the main model described
above, we fitted two-pollutant models, each of
which included adjustment for one of four gas-
eous pollutants. The association of PM concen-
tration with mortality was considered robust if
the effect estimates in the single-pollutant and
two-pollutant models were not significantly dif-
ferent, as determined with a paired z-test. An interactive map
is available at
Using the aforementioned two-stage approach, NEJM.org
we also performed regional analyses, with the
regions grouped according to WHO region and
according to the gross domestic product (GDP)
per capita (Table S2 in the Supplementary Ap-
pendix), and likelihood-ratio tests were used to
determine whether the differences between re-
gions in associations of PM with mortality were
significant. To further explore potential effect
modifications, we fit meta-regression models
with annual mean concentrations of PM and
copollutants, annual mean temperature, latitude
of locations, WHO region and region classified
according to the GDP per capita, rates of miss-
ing data on daily mortality and PM10 and PM2.5
concentrations, and GDP per capita.
To estimate the overall shape of the associa-
tions between PM10 and PM2.5 concentrations and
mortality at the global or country level, we plot-

n engl j med 381;8 nejm.org  August 22, 2019 707

The New England Journal of Medicine
Downloaded from nejm.org by Javier de Jesus Oquendo Contreras on September 10, 2019. For personal use only. No other uses without permission.
Copyright © 2019 Massachusetts Medical Society. All rights reserved.
 The n e w e ng l a n d j o u r na l of m e dic i n e

Mean Annual PM2.5

Concentration (µg/m3)
5–10
11–20
21–30
31–40
41–50
51–60
61–70
71–80
81–100
>100

Kilometers
0 2500 5000 10,000

Figure 2. Distribution of Cities with Data on PM2.5.

Shown is the geographic distribution of the 499 cities in the 16 countries and regions that had data on particulate matter with an aero-
dynamic diameter of 2.5 μm or less (PM2.5) and were included in the analysis. Also shown are the annual mean PM2.5 concentrations.
See the interactive map, available at NEJM.org.

ted concentration–response curves using the same
approach that was used in previous studies.16,19
In brief, we replaced the linear term of PM in the
main model with a B-spline function with two
knots at the 25th and 75th percentiles of the
mean PM concentrations across all cities.
We performed several sensitivity analyses. First,
in fitting the concentration–response curves, we
placed knots at different PM values. Second, we
tested the potential confounding effect of hu-
midity in cities that had available data on this
variable by comparing the results of models that
adjusted for humidity with the results of models
that did not in a paired z-test. Third, we re-
stricted the analyses to data available after the
year 2000.
We conducted all statistical analyses with
R software, version 3.3.1 (R Foundation for Statis-
tical Computing), using the mgcv package for fit-
ting main models and the rmeta package for
performing random-effect models. A P value of
less than 0.05 was considered to indicate statis-
tical significance. More details are presented in
the Methods section in the Supplementary Ap-
pendix.
R e sult s
Descriptive Analyses
The final analysis included 59.6 million deaths
from any cause or nonexternal causes, 20.1 mil-
lion deaths from cardiovascular diseases, and
5.6 million deaths from respiratory diseases (Ta-
ble S1 [nontrimmed data] and Table S3 [trimmed
data] in the Supplementary Appendix). On aver-
age, the annual mean concentration of PM10 in
598 cities was 56.0 μg per cubic meter (median,
44.3 μg per cubic meter [range, 11.0 to 295.0;
interquartile range, 37.9 to 70.1]), and the an-
nual mean concentration of PM2.5 in 499 cities
was 35.6 μg per cubic meter (median, 31.9 μg
per cubic meter [range, 4.1 to 116.9; interquartile
range, 21.5 to 43.5]). PM10 was strongly correlated
with PM2.5, with a mean Pearson correlation co-
efficient of 0.78. The mean Pearson correlation
coefficients between PM10 and gaseous pollutants

708 n engl j med 381;8 nejm.org  August 22, 2019

The New England Journal of Medicine

Downloaded from nejm.org by Javier de Jesus Oquendo Contreras on September 10, 2019. For personal use only. No other uses without permission.
Copyright © 2019 Massachusetts Medical Society. All rights reserved.
 Air Pollution and Mortality in 652 Cities
were 0.46 with nitrogen dioxide, 0.20 with ozone,
0.38 with sulfur dioxide, and 0.40 with carbon
monoxide. The corresponding coefficients be-
tween PM2.5 and gaseous pollutants were 0.48,
0.22, 0.40, and 0.45. Other descriptive statistics
and the correlations between daily mean PM
concentrations and weather variables are sum-
marized in the Results section in the Supple-
mentary Appendix.
Regression Analyses
The choice of a 2-day moving average for PM
concentration, which represents the average over
the same and previous day (lag 0 to 1 day), and
a 4-day moving average for temperature, which
represents the average of the same and previous
3 days (lag 0 to 3 days), generated the smallest
mean generalized cross-validation scores (Tables
S4 and S5 in the Supplementary Appendix). These
moving averages were then applied in subsequent
analyses. For both PM10 and PM2.5, the associa-
tions were significant on lag 0 day and then
attenuated substantially on lag 1 to 2 days; the
estimates of the associations were strongest on
lag 0 to 1 day (Table S4 in the Supplementary
Appendix).
Overall, we observed positive and significant
associations between PM10 and PM2.5 concentra-
tions and all-cause mortality (Table 1). In 598
cities that had data on PM10, an increase of 10 μg
per cubic meter in the PM10 concentration was
associated with an increase of 0.44% (95% con-
fidence interval [CI], 0.39 to 0.50) in a pooled
estimate of daily all-cause mortality. In 499 cities
that had data on PM2.5, the same increase in the
PM2.5 concentration was associated with an in-
crease of 0.68% (95% CI, 0.59 to 0.77) in a pooled
estimate of daily all-cause mortality. The country-
specific estimates of the percentage change in
daily all-cause mortality showed considerable
variations, ranging from 0.03% (for Colombia)
to 1.32% (for Australia) in association with a
10-μg-per-cubic-meter increase in PM10 concentra-
tion and ranging from 0.03% (for Portugal) to
2.54% (for Greece) in association with the same
increase in PM2.5 concentration. Estimates of the
effect in France, Estonia, and Switzerland were
close to the global median estimate of 0.46% in
association with PM10 concentration; estimates
of the effect in Switzerland and South Africa
were close to the global median estimate of 0.80%
in association with PM2.5 concentration.
n engl j med 381;8 nejm.org  August 22, 2019
The New England Journal of Medicine
Downloaded from nejm.org by Javier de Jesus Oquendo Contreras on September 10, 2019. For personal use only. No other uses without permission.
Copyright © 2019 Massachusetts Medical Society. All rights reserved.
In cause-specific analyses, an increase of 10 μg
per cubic meter in PM10 concentration (in 528
cities) was associated with an increase of 0.36%
(95% CI, 0.30 to 0.43) in daily cardiovascular
mortality and an increase of 0.47% (95% CI, 0.35
to 0.58) in daily respiratory mortality. The corre-
sponding increases in daily cardiovascular and
respiratory mortality for the same increase in PM2.5
concentration (in 488 cities) were 0.55% (95% CI,
0.45 to 0.66) and 0.74% (95% CI, 0.53 to 0.95%)
(Figs. S1 and S2 in the Supplementary Appen-
dix). In 445 cities that had data on both PM2.5
and PM10, the percentage increases in all-cause
mortality per 10-μg-per-cubic-meter increase in
PM2.5 concentration were larger than those with
the same increase in PM10 concentration, both in
the pooled results and in most country-specific
estimates (Fig. S3 in the Supplementary Appendix).
Regional analyses indicated differences be-
tween areas (Table S6 in the Supplementary Ap-
pendix), with higher estimates of the effect in
the region of the Americas and smaller estimates
in the Western Pacific region. We observed
stronger associations between PM10 and PM2.5
concentrations and all-cause mortality in loca-
tions with lower annual mean concentrations
of PM and higher annual mean temperatures
(P<0.001 for all comparisons); there was no sig-
nificant modification of the effect according to
annual mean concentrations of PM and copollut-
ants, latitude of location, WHO region and re-
gion classified according to the GDP per capita,
rates of missing data on daily mortality and PM10
and PM2.5 concentrations, and GDP per capita
(P>0.05 for all comparisons).
In two-pollutant models (Table 2), the magni-
tude (i.e., the size of the estimated effect) of the
associations of PM10 and PM2.5 concentrations on
lag 0 to 1 day with all-cause mortality decreased,
but all associations between PM and mortality
remained significant after adjustment for gaseous
pollutants. Notably, the estimates of the percent-
age change in mortality per 10-μg-per-cubic-
meter increase in PM10 concentration decreased
significantly after adjustment for nitrogen dioxide
(difference of 35%; P<0.001) and sulfur dioxide
(difference of 18%; P = 0.007). Similarly, the per-
centage change in mortality with the same in-
crease in PM2.5 concentration decreased by 36%
after adjustment for nitrogen dioxide (P<0.001)
and by 22% after adjustment for sulfur dioxide
(P = 0.007).
709
 The n e w e ng l a n d j o u r na l of m e dic i n e

Table 1. Percentage Change in All-Cause Mortality per 10-μg-per-Cubic-Meter Increase in 2-Day Moving Average
Concentrations of Inhalable Particulate Matter (PM10) and Fine Particulate Matter (PM2.5).*

Country or Region PM10 PM2.5

Cities with Cities with

Available Available
Data Pooled Estimate Data Pooled Estimate

no. % (95% CI) no. % (95% CI)

Australia 3 1.32 (0.22 to 2.44) 3   1.42 (–0.12 to 2.99)
Brazil 1 1.22 (0.97 to 1.47) 0 NA
Canada 13 0.76 (0.25 to 1.27) 25 1.70 (1.17 to 2.23)
Chile 4 0.33 (0.14 to 0.53) 4   0.27 (–0.68 to 1.23)
China 272 0.28 (0.22 to 0.34) 272 0.41 (0.32 to 0.50)
Colombia 1   0.03 (–0.34 to 0.39) 0 NA
Czech Republic 1   0.40 (–0.02 to 0.82) 0 NA
Estonia 4   0.46 (–0.69 to 1.63) 3   0.23 (–4.24 to 4.90)
Finland 1   0.07 (–0.51 to 0.65) 1   0.14 (–0.55 to 0.83)
France 18   0.46 (–0.15 to 1.07) 0 NA
Greece 1 0.53 (0.17 to 0.90) 1 2.54 (1.28 to 3.83)
Italy 18 0.65 (0.26 to 1.04) 0 NA
Japan 47 1.05 (0.78 to 1.31) 47 1.42 (1.05 to 1.81)
Mexico 8 0.67 (0.48 to 0.86) 3 1.29 (0.21 to 2.39)
Portugal 2   0.11 (–0.27 to 0.49) 1   0.03 (–1.14 to 1.21)
South Africa 6 0.41 (0.14 to 0.68) 5 0.80 (0.16 to 1.44)
South Korea 7 0.42 (0.27 to 0.58) 0 NA
Spain 45 0.87 (0.60 to 1.15) 19 1.96 (1.18 to 2.75)
Sweden 1   0.20 (–1.03 to 1.44) 1   0.08 (–1.44 to 1.62)
Switzerland 8   0.47 (–0.36 to 1.31) 4   0.79 (–0.96 to 2.58)
Taiwan 3   0.25 (–0.03 to 0.53) 3   0.62 (–0.39 to 1.64)
Thailand 19 0.61 (0.24 to 0.99) 0 NA
United Kingdom 15   0.06 (–0.36 to 0.48) 0 NA
United States 100 0.79 (0.60 to 0.98) 107 1.58 (1.28 to 1.88)
Total 598 0.44 (0.39 to 0.50) 499 0.68 (0.59 to 0.77)

* Pooled estimates represent the percentage changes in daily all-cause mortality per 10-μg-per-cubic-meter increase in
concentrations of particulate matter (PM) with an aerodynamic diameter of 10 μm or less (PM10) and PM with an aero-
dynamic diameter of 2.5 μm or less (PM2.5), as determined with the use of trimmed exposure data in which the highest
5% and lowest 5% of PM10 and PM2.5 measurements were excluded. NA denotes not available.

The concentration–response associations of
daily mean PM10 and PM2.5 concentrations with
all-cause mortality were positive, and the curves
showed a consistent increase with no discernible
thresholds (Fig. 3). The slopes for both curves
were steeper at concentrations lower than 40 μg
per cubic meter for PM10 and lower than 20 μg per
cubic meter for PM2.5. The slopes seemed to flat-
ten at high ranges. In addition, positive associa-
tions were still detectable at levels below most
global and regional air-quality guidelines or stan-
dards. Country-specific concentration–response
curves are provided in Figures S4 and S5 in the
Supplementary Appendix.
Sensitivity analyses confirmed these results.
The use of alternative knots did not substantially
change the shape of the concentration–response
curves, and adjustment for humidity resulted in

710 n engl j med 381;8 nejm.org  August 22, 2019

The New England Journal of Medicine

Downloaded from nejm.org by Javier de Jesus Oquendo Contreras on September 10, 2019. For personal use only. No other uses without permission.
Copyright © 2019 Massachusetts Medical Society. All rights reserved.
 Table 2. Percentage Change in All-Cause Mortality in Association with an Increase of 10 μg per Cubic Meter in the 2-Day Moving Average Concentrations of PM10 and PM2.5, with and
without Adjustment for Copollutants.*

Models PM10 PM2.5

Cities with P Value for Cities with P Value for

Available Data Pooled Estimate Difference Available Data Pooled Estimate Difference

no. % (95% CI) no. % (95% CI)

PM and ozone 559 0.90 487 0.75
Single-pollutant model of PM 0.43 (0.37–0.48) 0.68 (0.58–0.77)
Two-pollutant model of PM with adjustment 0.43 (0.37–0.49) 0.66 (0.56–0.77)
for ozone
PM and nitrogen dioxide 495 <0.001 466 <0.001
Single-pollutant model of PM 0.43 (0.37–0.49) 0.66 (0.57–0.76)
Two-pollutant model of PM with adjustment 0.28 (0.22–0.35) 0.42 (0.31–0.53)
for nitrogen dioxide
PM and sulfur dioxide 495 0.007 466 0.007
Single-pollutant model of PM 0.44 (0.38–0.50) 0.67 (0.57–0.76)
Two-pollutant model of PM with adjustment 0.36 (0.30–0.42) 0.52 (0.42–0.62)
for sulfur dioxide

n engl j med 381;8 nejm.org  August 22, 2019

The New England Journal of Medicine

PM and carbon monoxide 445 0.75 416 0.50
Air Pollution and Mortality in 652 Cities

Single-pollutant of PM 0.40 (0.34–0.46) 0.61 (0.51–0.71)

Two-pollutant model of PM with adjustment 0.39 (0.32–0.46) 0.57 (0.46–0.68)
for carbon monoxide

Copyright © 2019 Massachusetts Medical Society. All rights reserved.

* Pooled estimates are of the percentage change in daily all-cause mortality per 10-μg-cubic-meter increase in PM10 or PM2.5 concentration. The P value for difference was calculated by
evaluating a binary variable (with and without the adjustment for the copollutant) in a paired z-test with estimates from both single-pollutant and two-pollutant models. A P value of
less than 0.05 was considered to be statistically significant for the difference.

Downloaded from nejm.org by Javier de Jesus Oquendo Contreras on September 10, 2019. For personal use only. No other uses without permission.
711
 The n e w e ng l a n d j o u r na l of m e dic i n e

A PM10 B PM2.5
WHO AQG; WHO IT-3 WHO IT-2 WHO IT-1; WHO AQG US NAAQS WHO IT-3 WHO IT-2 WHO IT-1;
EU AQD US NAAQS; China AQS
China AQS
4 4
Percentage Difference in Mortality

Percentage Difference in Mortality

2 2

0 0

−2 −2

−4 −4

−6 −6
0 50 100 150 200 250 300 0 25 50 75 100 125 150
PM10 Concentration (µg/m3) PM2.5 Concentration (µg/m3)

Figure 3. Pooled Concentration–Response Curves.

Shown are the pooled concentration–response curves for the associations of 2-day moving average concentrations of PM10 (Panel A)
and PM2.5 (Panel B) with daily all-cause mortality. The y axis represents the percentage difference from the pooled mean effect (as de-
rived from the entire range of PM concentrations at each location) on mortality. Zero on the y axis represents the pooled mean effect,
and the portion of the curve below zero denotes a smaller estimate than the mean effect. The dashed lines represent the air-quality
guidelines or standards for 24-hour average concentrations of PM10 or PM2.5 according to the World Health Organization Air Quality
Guidelines (WHO AQG), WHO Interim Target 1 (IT-1), WHO Interim Target 2 (IT-2), WHO Interim Target 3 (IT-3), European Union Air
Quality Directive (EU AQD), U.S. National Ambient Air Quality Standard (NAAQS), and China Air Quality Standard (AQS).

no significant changes (Figs. S6 and S7 and Ta- crease of 0.44% in all-cause mortality per 10-μg-
ble S7 in the Supplementary Appendix). Finally, per-cubic-meter increase in PM10 concentration.
the analysis in which the subset of data since the The magnitude of the association is generally
year 2000 was used provided similar estimates. similar to previous findings in other multicity or
Estimates based on nontrimmed PM data are pro- multicountry studies.7-9,20 For example, the Air
vided in Table S8 in the Supplementary Appendix. Pollution and Health: A European and North
American Approach (APHENA) study reported
increases of 0.86%, 0.33%, and 0.29% in daily
Discussion
all-cause mortality in Canada, Europe, and the
Our study analyzed multisite data on air pollu- United States, respectively.9 The percentage in-
tion and mortality in 652 cities across different crease in mortality for the same increase in PM10
countries and regions, although most countries concentration was 0.77% in the Multicity Study
and cities were in the northern hemisphere. Be- of Air Pollution and Mortality in Latin America
cause the data from each city were analyzed ac- (ESCALA),8 0.55% in the Public Health and Air
cording to the same protocol, the estimate of Pollution in Asia (PAPA) study,7 and 0.19% in the
the percentage change in mortality per 10-μg- reanalysis of the U.S. National Morbidity Mortal-
per-cubic-meter increase in PM concentration ity Air Pollution Study (NMMAPS).21
was based on a large data set. This study also In the analysis of PM2.5, we observed an in-
provides the statistical power to examine the crease of 0.68% in all-cause mortality per 10-μg-
global concentration–response functions of par- per-cubic-meter increase in PM2.5 concentration.
ticulate air pollution at both low and high base- Our estimates were somewhat smaller than those
line levels. obtained from previous multicity studies and a
In the analysis of PM10, we observed an in- meta-analysis that used data mainly from devel-

712 n engl j med 381;8 nejm.org  August 22, 2019

The New England Journal of Medicine

Downloaded from nejm.org by Javier de Jesus Oquendo Contreras on September 10, 2019. For personal use only. No other uses without permission.
Copyright © 2019 Massachusetts Medical Society. All rights reserved.
 Air Pollution and Mortality in 652 Cities
oped countries.22,23 This difference may be inter-
preted as reflecting the nonlinearity of our
concentration–response curve, which indicated a
steeper slope at lower concentrations. In addition,
we found that the associations of mortality with
PM concentrations were slightly stronger with
PM2.5 than with PM10 in most countries and re-
gions, which added to the evidence that PM2.5
accounted for a larger proportion of the effects
of PM10 and PM2.5 combined.6 The stronger ef-
fects of PM2.5 may also be supported by the
abundant evidence that this particulate fraction
contains more small particles that can absorb
toxic components from the air and penetrate
deep into the lungs.24
The question of whether the observed asso-
ciations for PM were independent from other
pollutants is important for air-quality regulation
and health-risk assessment. In our data, although
the magnitude of the associations for PM10 and
PM2.5 decreased in two-pollutant models, the as-
sociations for both remained significant, a find-
ing that provides evidence of the independent
health effects of PM. It is notable that the esti-
mates of the percentage change in mortality per
10-μg-per-cubic-meter increase in PM10 and PM2.5
concentrations decreased more after adjustment
for nitrogen dioxide and sulfur dioxide than af-
ter adjustment for ozone and carbon monoxide,
a finding that may be interpreted as reflecting
closer correlations of PM with nitrogen dioxide
and sulfur dioxide caused by similar sources and
seasonal patterns.
In accordance with the findings from the
majority of previous studies, the concentration–
response curves between PM concentration and
daily mortality derived from this global study
showed a consistent increase without evidence of
a threshold.16,19,22 In both curves, the percentage
increase in mortality per unit change in PM con-
centration seemed to be smaller (i.e., the con-
centration–response curves seemed to flatten) at
high ranges of daily mean PM concentration. This
potential saturation effect may be explained by
smaller effects of changes in daily mean PM
concentration in cities with higher baseline levels
of PM, as suggested in our meta-regression analy-
ses. Furthermore, the higher proportion of young
people in developing countries may decrease
population susceptibility to PM, and less out-
door activity during days with high pollution
levels may decrease exposure. Nevertheless, the
n engl j med 381;8 nejm.org  August 22, 2019
The New England Journal of Medicine
Downloaded from nejm.org by Javier de Jesus Oquendo Contreras on September 10, 2019. For personal use only. No other uses without permission.
Copyright © 2019 Massachusetts Medical Society. All rights reserved.
concentrations of PM below the current air-
quality guidelines and standards may still be
hazardous to public health. However, associa-
tions estimated for extreme PM concentrations
are characterized by wider confidence intervals,
with greater uncertainty about the actual mortal-
ity risk at such values. We should also be cautious
about the uncertainty in the concentration–­
response curves, because they were pooled from
cities or countries with diverse PM ranges and
varying population susceptibility and data qual-
ity and representability.
We found significant evidence of spatial hetero-
geneity in the associations between PM concen-
tration and daily mortality across countries and
regions. A number of factors could contribute to
this variability, including different PM compo-
nents, long-term air pollution levels, population
susceptibility, and different lengths of study peri-
ods. We also found that higher annual mean
concentrations of PM10 and PM2.5 were accompa-
nied by weaker associations with daily mortality,
a finding that has been reported in previous
studies.16,25 The possible adaptive response to PM
in populations living in areas with higher long-
term exposure to PM may lead to smaller esti-
mate-per-unit changes in exposure. In addition,
we identified stronger associations of PM with
mortality in regions with higher GDP per capita,
which may also be in relation to lower long-term
air pollution levels (Pearson coefficient, –0.68
for PM10 and –0.74 for PM2.5) and decreased
population susceptibility due to higher socioeco-
nomic status.26 The estimates of the association
between PM and mortality in some countries
(e.g., France, Finland, Sweden, and the United
Kingdom) were smaller and not significant.
These countries had fewer cities included and
shorter periods evaluated, which may increase
the statistical uncertainty in the estimation of
the effect. Furthermore, these countries are gen-
erally located in areas with a low annual mean
temperature, which may decrease the association
between PM and mortality, as shown in meta-
regression analyses. More interpretations on this
issue are provided in the Discussion section in
the Supplementary Appendix.
This study has several limitations. First, al-
though the analysis included 24 major countries
and regions on six continents, our findings can-
not be interpreted as fully globally representative
because the 652 cities were mainly located in
713
 The n e w e ng l a n d j o u r na l of m e dic i n e

East Asia, Europe, and North America, with a concentrations, and the associations were still
smaller number of cities in Latin America and detectable at concentrations below the current
Africa. Second, we relied on fixed-site environ- air-quality guidelines and regulatory limits.
mental measurements, which could introduce ex- Supported by the National Natural Science Foundation of China
posure misclassification. Third, diagnostic or cod- (grants 91843302 and 91643205, to Dr. Kan); the China Medical
Board Collaborating Program (grant 16-250); the Medical Re-
ing errors in health data are also inevitable in search Council, United Kingdom (grants MR/R013349/1 and
such a global study that spans multiple decades; MR/M022625/1, to Drs. Gasparrini, Sera, and Vicedo-Cabrera);
the effects of these errors on our results are dif- the Career Development Fellowship of the Australian National
Health and Medical Research Council (grants APP1107107 and
ficult to evaluate, which presumably makes the APP1163693, to Dr. Guo); the Ministry of Education of Spain
estimates of the effects on cause-specific mortal- (grant PRX17/00705, to Dr. Tobias); the National Plan for I+D+I
ity less reliable than those of effects on all-cause (grant PI15/00515), cofunded by the Instituto de Salud Carlos III
Directorate General for Evaluation and the European Regional
mortality. Fourth, there are some missing data, Development Fund (FEDER); the Global Research Laboratory
but their influence on our estimates was not (grant K21004000001-10A0500-00710 from the National Re-
substantial (see the Discussion section and Table search Foundation of Korea, funded by the Ministry of Science,
Information and Communication Technologies, to Dr. Kim); the
S9 in the Supplementary Appendix). Academy of Finland (grants 310372 and 310373, to Drs. Jaakkola
Our multicountry time-series analysis provides and Ryti); the Estonian Ministry of Education and Research
evidence on positive associations between short- (grant IUT34-17, to Drs. Orru and Indermitte); the Czech Science
Foundation (grant 18-22125S, to Drs. Kyselý and Urban); and a
term exposure to PM10 and PM2.5 and daily all- Professional Services Agreement with the Health Effects Insti-
cause, cardiovascular, and respiratory mortality. tute, United States (to Dr. Cohen).
This study indicated independent associations of No potential conflict of interest relevant to this article was
reported.
PM10 and PM2.5 concentrations with daily mortal- Disclosure forms provided by the authors are available with
ity after adjustment for gaseous pollutants. Fur- the full text of this article at NEJM.org.
ther, concentration–response curves for the ef- We thank Benjawan Tawatsupa and Kornwipa Punnasiri for
providing environmental data for Thailand and Fiorella Acquaotta
fects of PM on mortality showed a consistent at the University of Turin for providing the temperature data for
increase, with flattening of the slopes at higher South Africa.

Appendix
The authors’ full names and academic degrees are as follows: Cong Liu, M.S., Renjie Chen, Ph.D., Francesco Sera, Ph.D., Ana M.
Vicedo‑Cabrera, Ph.D., Yuming Guo, Ph.D., Shilu Tong, Ph.D., Micheline S.Z.S. Coelho, Ph.D., Paulo H.N. Saldiva, Ph.D., Eric Lavigne,
Ph.D., Patricia Matus, Ph.D., Nicolas Valdes Ortega, M.Sc., Samuel Osorio Garcia, Ph.D., Mathilde Pascal, Ph.D., Massimo Stafoggia,
Ph.D., Matteo Scortichini, Ph.D., Masahiro Hashizume, Ph.D., Yasushi Honda, Ph.D., Magali Hurtado‑Díaz, Ph.D., Julio Cruz, Ph.D.,
Baltazar Nunes, Ph.D., João P. Teixeira, Ph.D., Ho Kim, Ph.D., Aurelio Tobias, Ph.D., Carmen Íñiguez, Ph.D., Bertil Forsberg, Ph.D.,
Christofer Åström, Ph.D., Martina S. Ragettli, Ph.D., Yue‑Leon Guo, Ph.D., Bing‑Yu Chen, Ph.D., Michelle L. Bell, Ph.D., Caradee Y.
Wright, Ph.D., Noah Scovronick, Ph.D., Rebecca M. Garland, Ph.D., Ai Milojevic, Ph.D., Jan Kyselý, Ph.D., Aleš Urban, Ph.D., Hans
Orru, Ph.D., Ene Indermitte, Ph.D., Jouni J.K. Jaakkola, Ph.D., Niilo R.I. Ryti, Ph.D., Klea Katsouyanni, Ph.D., Antonis Analitis, Ph.D.,
Antonella Zanobetti, Ph.D., Joel Schwartz, Ph.D., Jianmin Chen, Ph.D., Tangchun Wu, Ph.D., Aaron Cohen, D.Sc., Antonio Gasparrini,
Ph.D., and Haidong Kan, Ph.D.
The authors’ affiliations are as follows: the School of Public Health, Key Laboratory of Public Health Safety of the Ministry of Educa-
tion and National Health Commission Key Laboratory of Health Technology Assessment (C.L., R.C., H. Kan), the Department of Envi-
ronmental Science and Engineering (J. Chen), Shanghai Key Laboratory of Atmospheric Particle Pollution and Prevention (R.C., H. Kan),
Fudan University, Shanghai Children’s Medical Center, Shanghai Jiao Tong University School of Medicine (S.T.), and Children’s Hos-
pital of Fudan University, National Center for Children’s Health (H. Kan), Shanghai, the School of Public Health and Management,
Binzhou Medical University, Yantai (Y.G.), the School of Public Health, Institute of Environment and Population Health, Anhui Medical
University, Hefei (S.T.), and the Key Laboratory of Environment and Health, Ministry of Education, and State Key Laboratory of Envi-
ronmental Health (Incubating), School of Public Health, Tongji Medical College, Huazhong University of Science and Technology,
Wuhan (T.W.) — all in China; the Department of Public Health, Environments and Society (F.S., A.M.V.-C., A.M., A.G.) and the Centre
for Statistical Methodology (A.G.), London School of Hygiene and Tropical Medicine, and the School of Population Health and Envi-
ronmental Sciences, King’s College London (K.K.), London; the Department of Epidemiology and Preventive Medicine, School of
Public Health and Preventive Medicine, Monash University, Melbourne, VIC (Y.G.), and the School of Public Health and Social Work,
Queensland University of Technology, Brisbane (S.T.) — both in Australia; the Institute of Advanced Studies, University of São Paulo,
São Paulo (M.S.Z.S.C., P.H.N.S.); the Air Health Science Division, Health Canada, and the School of Epidemiology and Public Health,
University of Ottawa, Ottawa, ON (E.L.); the Department of Public Health (P.M.) and the School of Nursing and Obstetrics (N.V.O.),
Universidad de los Andes, Santiago, Chile; Hospital Vista Hermosa, Bogota, Colombia (S.O.G.); Santé Publique France, French Na-
tional Public Health Agency, Saint Maurice, France (M.P.); the Department of Epidemiology, Lazio Regional Health Service–ASL Roma 1,
Rome (M. Stafoggia, M. Scortichini); Karolinska Institute, Institute of Environmental Medicine, Stockholm (M. Stafoggia), and the
Department of Public Health and Clinical Medicine, Umeå University, Umeå (B.F., C.Å., H.O.) — both in Sweden; the Department of
Pediatric Infectious Diseases, Institute of Tropical Medicine, Nagasaki University, Nagasaki (M.H.), and the Faculty of Health and Sport
Sciences, University of Tsukuba, Tsukuba (Y.H.) — both in Japan; the Department of Environmental Health, National Institute of Pub-

714 n engl j med 381;8 nejm.org  August 22, 2019

The New England Journal of Medicine

Downloaded from nejm.org by Javier de Jesus Oquendo Contreras on September 10, 2019. For personal use only. No other uses without permission.
Copyright © 2019 Massachusetts Medical Society. All rights reserved.
 Air Pollution and Mortality in 652 Cities

lic Health, Cuernavaca, Mexico (M.H.-D., J. Cruz); the Department of Epidemiology, Instituto Nacional de Saúde Dr. Ricardo Jorge,
Lisbon (B.N., J.P.T.), and the Epidemiology Research Unit–Instituto de Saúde Pública, Universidade do Porto, Porto (J.P.T.) — both in
Portugal; the Department of Public Health Science, Graduate School of Public Health and Institute of Health and Environment, Seoul
National University, Seoul, South Korea (H. Kim); the Institute of Environmental Assessment and Water Research, Spanish Council for
Scientific Research, Barcelona (A.T.), and the Department of Statistics and Computational Research, University of Valencia Environmental
Health Joint Research Unit Fundación para el Fomento de la Investigación Sanitaria y Biomédica de la Comunitat Valenciana–Universitat
de València–Universitat Jaume I de Castellón Biomedical Research Center Network for Epidemiology and Public Health, Valencia (C.I.)
— both in Spain; the Swiss Tropical and Public Health Institute and the University of Basel, Basel, Switzerland (M.S.R.); Environmental
and Occupational Medicine, National Taiwan University (Y.-L.G., B.-Y.C.), and the College of Medicine and National Taiwan University
Hospital (Y.-L.G.), Taipei City; the School of Forestry and Environmental Studies, Yale University, New Haven, CT (M.L.B.); the Environ-
ment and Health Research Unit, South African Medical Research Council (C.Y.W.), the Department of Geography, Geo-informatics,
and Meteorology, University of Pretoria (C.Y.W., R.M.G.), and the Natural Resources and the Environment Unit, Council for Scientific
and Industrial Research (R.M.G.), Pretoria, and the Unit for Environmental Sciences and Management, North-West University, Potchef-
stroom (R.M.G.) — all in South Africa; the Department of Environmental Health, Rollins School of Public Health, Emory University,
Atlanta (N.S.); the Institute of Atmospheric Physics, Czech Academy of Sciences, (J.K., A.U.), and the Faculty of Environmental Sci-
ences (J.K.), Czech University of Life Sciences, Prague, Czech Republic; the Institute of Family Medicine and Public Health, University
of Tartu, Tartu, Estonia (H.O., E.I.); the Center for Environmental and Respiratory Health Research, University of Oulu, Medical Re-
search Center Oulu, and Oulu University Hospital and University of Oulu, Oulu, Finland (J.J.K.J., N.R.I.R.); the Department of Hygiene,
Epidemiology and Medical Statistics, School of Medicine, National and Kapodistrian University of Athens, Athens (K.K., A.A.); the
Department of Environmental Health, Harvard T.H. Chan School of Public Health (A.Z., J.S.), and the Health Effects Institute (A.C.),
Boston; and the Institute for Health Metrics and Evaluation, University of Washington, Seattle (A.C.).

References
1. Rückerl R, Schneider A, Breitner S,
Cyrys J, Peters A. Health effects of par-
ticulate air pollution: a review of epide-
miological evidence. Inhal Toxicol 2011;​
23:​555-92.
2. Brunekreef B, Holgate ST. Air pollu-
tion and health. Lancet 2002;​360:​1233-42.
3. Héroux ME, Anderson HR, Atkinson
R, et al. Quantifying the health impacts of
ambient air pollutants: recommendations
of a WHO/Europe project. Int J Public
Health 2015;​60:​619-27.
4. WHO air quality guidelines for par-
ticulate matter, ozone, nitrogen dioxide
and sulfur dioxide:​global update 2005:​
summary of risk assessment. Geneva:​
World Health Organization, 2006.
5. Dai L, Zanobetti A, Koutrakis P,
Schwartz JD. Associations of fine particu-
late matter species with mortality in the
United States: a multicity time-series analy-
sis. Environ Health Perspect 2014;​122:​837-
42.
6. Lu F, Xu D, Cheng Y, et al. Systematic
review and meta-analysis of the adverse
health effects of ambient PM2.5 and PM10
pollution in the Chinese population. Envi-
ron Res 2015;​136:​196-204.
7. Wong CM, Vichit-Vadakan N, Kan H,
Qian Z. Public Health and Air Pollution in
Asia (PAPA): a multicity study of short-
term effects of air pollution on mortality.
Environ Health Perspect 2008;​116:​1195-
202.
8. Romieu I, Gouveia N, Cifuentes LA,
et al. Multicity study of air pollution and
mortality in Latin America (the ESCALA
study). Res Rep Health Eff Inst 2012;​171:​
5-86.
9. Katsouyanni K, Samet JM, Anderson
HR, et al. Air pollution and health: a Eu-
ropean and North American approach
(APHENA). Res Rep Health Eff Inst 2009;​
142:​5-90.
10. Gasparrini A, Guo Y, Hashizume M,
et al. Temporal variation in heat-mortality
associations: a multicountry study. Envi-
ron Health Perspect 2015;​123:​1200-7.
11. Guo Y, Gasparrini A, Armstrong B,
et al. Global variation in the effects of am-
bient temperature on mortality: a system-
atic evaluation. Epidemiology 2014;​ 25:​
781-9.
12. Gasparrini A, Guo Y, Hashizume M,
et al. Mortality risk attributable to high
and low ambient temperature: a multi-
country observational study. Lancet 2015;​
386:​369-75.
13. International statistical classification
of diseases and related health problems.
Geneva:​World Health Organization, 2004.
14. Samet JM, Dominici F, Curriero FC,
Coursac I, Zeger SL. Fine particulate air
pollution and mortality in 20 U.S. cities,
1987–1994. N Engl J Med 2000;​343:​1742-9.
15. Bell ML, Dominici F, Samet JM. A meta-
analysis of time-series studies of ozone
and mortality with comparison to the Na-
tional Morbidity, Mortality, and Air Pollu-
tion Study. Epidemiology 2005;​16:​436-45.
16. Chen R, Yin P, Meng X, et al. Fine par-
ticulate air pollution and daily mortality:
a nationwide analysis in 272 Chinese cities.
Am J Respir Crit Care Med 2017;​196:​73-81.
17. Dominici F, Peng RD, Bell ML, et al.
Fine particulate air pollution and hospital
admission for cardiovascular and respira-
tory diseases. JAMA 2006;​295:​1127-34.
18. Shah AS, Langrish JP, Nair H, et al.
Global association of air pollution and
heart failure: a systematic review and
meta-analysis. Lancet 2013;​382:​1039-48.
19. Samoli E, Analitis A, Touloumi G, et al.
Estimating the exposure-response rela-
tionships between particulate matter and
mortality within the APHEA multicity
project. Environ Health Perspect 2005;​
113:​88-95.
20. Samoli E, Peng R, Ramsay T, et al.
Acute effects of ambient particulate mat-
ter on mortality in Europe and North
America: results from the APHENA study.
Environ Health Perspect 2008;​116:​1480-6.
21. Dominici F, Peng RD, Zeger SL, White
RH, Samet JM. Particulate air pollution
and mortality in the United States: did
the risks change from 1987 to 2000? Am
J Epidemiol 2007;​166:​880-8.
22. Di Q, Dai L, Wang Y, et al. Association
of short-term exposure to air pollution
with mortality in older adults. JAMA 2017;​
318:​2446-56.
23. Atkinson RW, Kang S, Anderson HR,
Mills IC, Walton HA. Epidemiological
time series studies of PM2.5 and daily mor-
tality and hospital admissions: a systemat-
ic review and meta-analysis. Thorax 2014;​
69:​660-5.
24. Kim KH, Kabir E, Kabir S. A review on
the human health impact of airborne par-
ticulate matter. Environ Int 2015;​74:​136-43.
25. Chen R, Kan H, Chen B, et al. Asso-
ciation of particulate air pollution with
daily mortality: the China Air Pollution
and Health Effects Study. Am J Epidemiol
2012;​175:​1173-81.
26. Martins MCH, Fatigati FL, Véspoli TC,
et al. Influence of socioeconomic condi-
tions on air pollution adverse health ef-
fects in elderly people: an analysis of six
regions in São Paulo, Brazil. J Epidemiol
Community Health 2004;​58:​41-6.
Copyright © 2019 Massachusetts Medical Society.

n engl j med 381;8 nejm.org  August 22, 2019 715

The New England Journal of Medicine
Downloaded from nejm.org by Javier de Jesus Oquendo Contreras on September 10, 2019. For personal use only. No other uses without permission.
Copyright © 2019 Massachusetts Medical Society. All rights reserved.